HiddenLayer Introduces Taxonomy for Adversarial Prompt Engineering

Security firm HiddenLayer has unveiled a structured taxonomy for adversarial prompt engineering, detailed in a company blog post. The framework organizes malicious prompt behaviors targeting large language models into hierarchical layers of tactics, techniques, and prompts, supplemented by defined attacker objectives such as data theft or reputation harm.

The taxonomy is designed to give researchers, red teamers, and AI security professionals a shared language for classifying and defending against prompt-based attacks. It builds on cybersecurity’s Tactics, Techniques, and Procedures (TTP) model, adding an “objectives” layer to separate intent from method. This structure enables teams to analyze adversarial behaviors systematically and align defensive measures accordingly.

HiddenLayer emphasized that the taxonomy is a work in progress and invites community input to refine and expand it as new techniques emerge. The interactive version of the taxonomy is available online for exploration and contribution.

The company will further explore this topic in its upcoming webinar, “Adversarial Prompt Engineering: Executive Insights and Red Team Lessons”, scheduled for October 29, 2025. The session will feature experts discussing the evolving threat landscape, organizational exposure assessment, and the role of red-teaming in strengthening AI resilience.